In [1]:
import matplotlib.pyplot as plt
from skimage import io
from matplotlib import cm
from mpl_toolkits.axes_grid1 import ImageGrid
import numpy as np
import os, PIL, pathlib, math
from imgaug import augmenters as iaa
from PIL import Image
import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import layers
from tensorflow.keras.models import Sequential
from tensorflow.keras.callbacks import ModelCheckpoint, EarlyStopping
from tensorflow.keras.preprocessing.image import ImageDataGenerator
%matplotlib inline
strategy = tf.distribute.MirroredStrategy()
print('Number of devices: {}'.format(strategy.num_replicas_in_sync))
def show_grid(image_list,nrows,ncols,label_list=None,show_labels=False,savename=None,figsize=(10,10),showaxis='off'):
if type(image_list) is not list:
if(image_list.shape[-1]==1):
image_list = [image_list[i,:,:,0] for i in range(image_list.shape[0])]
elif(image_list.shape[-1]==3):
image_list = [image_list[i,:,:,:] for i in range(image_list.shape[0])]
fig = plt.figure(None, figsize,frameon=False)
grid = ImageGrid(fig, 111, # similar to subplot(111)
nrows_ncols=(nrows, ncols), # creates 2x2 grid of axes
axes_pad=0.3, # pad between axes in inch.
share_all=True,
)
for i in range(nrows*ncols):
ax = grid[i]
ax.imshow((image_list[i]*255).astype(np.uint8),cmap='Greys_r') # The AxesGrid object work as a list of axes.
ax.axis('off')
if show_labels:
ax.set_title(class_mapping[y_int[i]])
if savename != None:
plt.savefig(savename,bbox_inches='tight')
# Defining Image augmentors
data_augmentation = keras.Sequential([
layers.experimental.preprocessing.RandomRotation(0.2),
layers.experimental.preprocessing.RandomTranslation(
0.2, 0.2, fill_mode='reflect', interpolation='bilinear'),
])
aug3 = iaa.MotionBlur(k=5, angle=[-90, 90])
aug4 = iaa.CoarseSaltAndPepper((0,0.02), size_px=(2, 8))
aug5 = iaa.AdditiveGaussianNoise(scale=(0, 0.05*255))
aug6 = iaa.Sharpen(alpha=(0,0.05), lightness=1.0)
aug7 = iaa.Dropout(p=(0, 0.05))
aug8 = iaa.LinearContrast((0.9, 1.0))
def add_aug(image):
#image = aug3.augment_image(image)
#image = aug4.augment_image(image)
#image = aug5.augment_image(image)
#image = aug6.augment_image(image)
#image = aug7.augment_image(image)
#image = aug8.augment_image(image)
return image
# Defining the three models in functions
def simpleModel():
model = Sequential([
data_augmentation,
#layers.experimental.preprocessing.Rescaling(1./255),
layers.Conv2D(16, 3, padding='same', activation='relu'),
layers.MaxPooling2D(),
layers.Conv2D(32, 3, padding='same', activation='relu'),
layers.MaxPooling2D(),
layers.Conv2D(64, 3, padding='same', activation='relu'),
layers.MaxPooling2D(),
layers.Dropout(0.2),
layers.Flatten(),
layers.Dense(128, activation='relu'),
layers.Dense(num_classes)
])
model.compile(optimizer='adam',
loss=tf.keras.losses.CategoricalCrossentropy(from_logits=True),
metrics=['accuracy'])
return model
def AlexNet(): #batch 32, 224, 0.002, 0.01, True
model = Sequential([
data_augmentation,
#layers.experimental.preprocessing.Rescaling(1./255),
layers.Conv2D(96, 11, strides=4, padding='valid', activation='relu'),
layers.BatchNormalization(),
layers.MaxPooling2D(pool_size=3, strides=2),
layers.Conv2D(256, 5, strides=1, padding='same', activation='relu'),
layers.BatchNormalization(),
layers.MaxPooling2D(pool_size=3, strides=2),
layers.Conv2D(384, 3, strides=1, padding='same', activation='relu'),
layers.BatchNormalization(),
layers.Conv2D(384, 3, strides=1, padding='same', activation='relu'),
layers.BatchNormalization(),
layers.Conv2D(256, 3, strides=1, padding='same', activation='relu'),
layers.BatchNormalization(),
layers.MaxPooling2D(pool_size=3, strides=2),
layers.Flatten(),
layers.Dense(4096, activation='relu'),
layers.Dropout(0.5),
layers.Dense(4096, activation='relu'),
layers.Dropout(0.5),
layers.Dense(num_classes)
])
opt = tf.keras.optimizers.SGD(learning_rate=0.001, momentum=0.01, nesterov=True)
model.compile(optimizer=opt,
loss=tf.keras.losses.CategoricalCrossentropy(from_logits=True),
metrics=['accuracy'])
return model
def VGG16(): #batch 64, 224, 0.002, 0.01, True
model = Sequential([
data_augmentation,
#layers.experimental.preprocessing.Rescaling(1./255),
layers.Conv2D(64, 3, strides=1, padding='same', activation='relu'),
layers.Conv2D(64, 3, strides=1, padding='same', activation='relu'),
layers.MaxPooling2D(pool_size=2), # stride defaults to pool_size which is OK
layers.Conv2D(128, 3, strides=1, padding='same', activation='relu'),
layers.Conv2D(128, 3, strides=1, padding='same', activation='relu'),
layers.MaxPooling2D(pool_size=2), # stride defaults to pool_size which is OK
layers.Conv2D(256, 3, strides=1, padding='same', activation='relu'),
layers.Conv2D(256, 3, strides=1, padding='same', activation='relu'),
layers.Conv2D(256, 3, strides=1, padding='same', activation='relu'),
layers.MaxPooling2D(pool_size=2), # stride defaults to pool_size which is OK
layers.Conv2D(512, 3, strides=1, padding='same', activation='relu'),
layers.Conv2D(512, 3, strides=1, padding='same', activation='relu'),
layers.Conv2D(512, 3, strides=1, padding='same', activation='relu'),
layers.MaxPooling2D(pool_size=2), # stride defaults to pool_size which is OK
layers.Conv2D(512, 3, strides=1, padding='same', activation='relu'),
layers.Conv2D(512, 3, strides=1, padding='same', activation='relu'),
layers.Conv2D(512, 3, strides=1, padding='same', activation='relu'),
layers.MaxPooling2D(pool_size=2), # stride defaults to pool_size which is OK
layers.Flatten(),
layers.Dense(4096, activation='relu'),
layers.Dense(4096, activation='relu'),
layers.Dense(num_classes)
])
opt = tf.keras.optimizers.SGD(learning_rate=0.02, momentum=0.2, nesterov=True)
model.compile(optimizer=opt,
loss=tf.keras.losses.CategoricalCrossentropy(from_logits=True),
metrics=['accuracy'])
return model
num_classes = 7
batch_size = 64
img_height = 224 # 180, 227, 224
img_width = 224 # 180, 227, 224
data_dir = pathlib.Path('takaFrames_dataset/')
#image_count = len(list(data_dir.glob('*/*.jpg')))
#print(image_count)
print("\n")
datagen_args = dict(rescale=1./255, validation_split=0.1, preprocessing_function=add_aug)
datagen = ImageDataGenerator(**datagen_args)
train_ds = datagen.flow_from_directory(
data_dir,
subset="training",
target_size=(img_height, img_width),
batch_size=batch_size)
print("\n")
val_ds = datagen.flow_from_directory(
data_dir,
subset="validation",
target_size=(img_height, img_width),
batch_size=batch_size)
print("\n")
print(train_ds.class_indices)
class_mapping = {v:k for k,v in train_ds.class_indices.items()}
x,y = next(train_ds)
print('x: ',type(x))
print('y: ',type(y))
print('x: ',x.shape)
print('y: ',y.shape)
print("\n")
y_int = np.argmax(y,axis=-1)
show_grid(x,4,8,label_list=y_int,show_labels=True,figsize=(20,10))
INFO:tensorflow:Using MirroredStrategy with devices ('/job:localhost/replica:0/task:0/device:GPU:0', '/job:localhost/replica:0/task:0/device:GPU:1', '/job:localhost/replica:0/task:0/device:GPU:2', '/job:localhost/replica:0/task:0/device:GPU:3')
Number of devices: 4
Found 43694 images belonging to 7 classes.
Found 4851 images belonging to 7 classes.
{'Children': 0, 'Cosmetics': 1, 'FashionAccessories': 2, 'Household': 3, 'Ladies': 4, 'Men': 5, 'Sports': 6}
x: <class 'numpy.ndarray'>
y: <class 'numpy.ndarray'>
x: (64, 224, 224, 3)
y: (64, 7)
